Scattering Source Determination Using Deep Learning for 2-D Scalar Wave Propagation

Simulations focusing on wave propagation and inverse problems involving the estimation of a scattering source have been conducted for a long time. As a method for estimating a scattering source of wave propagation, there are known techniques such as time-reversal methods that utilize simulation results and inverse scattering analysis methods based on the Born approximation. However, these methods are mathematically complex and require much computation time. Therefore, in this study, we aim to develop a scattering source estimation method using deep learning, which has been receiving increasing attention in recent years. However, the waveforms used for this scattering source estimation are generated using simulations. In this paper, we first simulate the 2-D scalar waves from the scattering source using the convolution quadrature time-domain boundary element method (CQBEM). The received waveforms at observation points are transformed into image data. These image data are utilized for the deep learning to estimate the actual position of the scattering source. As numerical examples, some unlearned waveforms by a scattering source are given to the created deep learning model and the position and size of the scattering source are estimated to verify the proposed method.